Older CUDA note
CUDA Thread Indexing Explained.
A compact replacement for an older technical note on global CUDA thread indexing across common grid and block layouts, preserved because developers still find the old article through outside references.
Preserved technical context
Why developers still find this topic.
Old citations to the article still appear in developer material. This page keeps a stable, edited URL for the note and points readers back to the broader GPU-computing work: cognitive robotics, iCub experiments, Aquila, and CUDA-era neural-network acceleration.
1DGlobal index in one dimension
For a one-dimensional launch, the usual global index is the block offset plus the local thread index: blockIdx.x * blockDim.x + threadIdx.x. The point is simple but essential: each GPU thread needs a unique coordinate in the work it is processing.
GPU research context
2D / 3DFlattening grids and blocks
The same idea extends by flattening x, y, and z coordinates into a linear index based on grid and block dimensions. That mapping is how image pixels, matrix elements, neural-network weights, or simulated agents can be distributed across thousands of threads.
NVIDIA CUDA spotlight
Why it matteredRobotics made compute practical
CUDA indexing was a small but necessary part of making larger neural-network and robotics experiments feasible on GPU hardware. In the PhD work, this kind of low-level parallel thinking sat underneath higher-level iCub, Aquila, MTRNN, and action-learning systems.
Sources
Archive role
A technical waypoint from the old site.
The original post belonged to a period when GPU programming was moving from specialist practice into broader robotics and AI experimentation. The current page keeps the useful developer context, fixes the grammar and structure, and connects the note to the work it supported.
AquilaGPU-accelerated cognitive robotics
Aquila and the iCub experiments turned CUDA from a programming topic into a practical research tool for action learning and embodied cognition.
Aquila archive
ThesisGPU Computing for Cognitive Robotics
The thesis and related papers show the larger research context: neural networks, humanoid robotics, and massively parallel computation under real experimental constraints.
Open thesis
SourcesPublic sources
For the full record, use the Sources page rather than old blog fragments. It carries the thesis, papers, NVIDIA references, archive PDFs, and external publication records.
Open Sources